1. Field of the Invention
The present invention relates to signal transmission and detection, and in particular to adaptive signal equalization for compensation of signal distortions caused by signal dispersion and nonlinearities within signal transmission media.
2. Description of the Related Art
Signal processing architectures for mitigation of different kinds of channel impairments and/or timing recovery and synchronization functions as used for communications transmission and/or storage systems can be divided into two categories: (1) discrete-time architecture (this architecture uses a sampled approach to convert the input continuous-time, analog waveform into a discrete signal and is commonly used in current systems; typically, a high resolution analog-to-digital converter, which follows the analog anti-aliasing filter, is used as the sampler at the analog front end); and (2) continuous-time architecture (this architecture is an analog continuous-time approach which directly processes the incoming analog waveform for mitigating channel impairments or timing recovery functions while remaining in the continuous time domain until the final data bit stream is generated).
In continuous-time and discrete-time signal processing architectures for adaptive equalization with LMS-based adaptation, the filter tap coefficients may be adapted based on a continuous-time or discrete-time basis based on the correlation of the error signal (as computed as the difference between the slicer output and time-aligned slicer input) and the corresponding time-aligned data signal input to the tap. It is then necessary to time-align the error signal and data signal and reduce any performance degradation that would otherwise arise. It is also commonly a design parameter to split the precursor and postcursor taps on the feedforward filter, whether operating alone or with decision feedback. Thus, a method which can explicitly control this within the adaptive equalizer would be desirable.
Fractional-spaced feedforward filters have commonly been used either as stand-alone linear equalizers or in combination with decision feedback. The adaptation technique for the tap coefficients implicitly assume independence in the adaptation of the successive tap coefficients, which has been based on minimizing the mean squared error (as computed as the difference between the slicer input, or pre-slice, signal and slicer output, or post-slice, signal). This adaptation technique is referred to as least mean square error (LMSE) or minimum mean square error (MMSE) adaptation. It can be shown that the LMSE adaptation for both fractional feedforward or symbol spaced feedback at iteration k+1 reduces to the following coefficient update equations:
                              c          _                =                              ∫            0            t                    ⁢                                    μ              ·                              e                ⁡                                  (                  t                  )                                                      ⁢                                          s                _                            ⁡                              (                t                )                                      ⁢                          ⅆ              t                                                                                      (                  continuous          ⁢                      -                    ⁢          time          ⁢                                          ⁢          adaptation          ⁢                                          ⁢          case                )            where c is the tap coefficient vector and e(t) the corresponding error (between delay-aligned slicer input and output), s is the vector with components as the input waveform to the corresponding tap mixer and time-aligned with the error signal appropriately and μ is a constant and is an adaptation parameter. Specifically, we have
      c    i    =            ∫      0      t        ⁢                  μ        ·                  e          ⁡                      (            t            )                          ·                  s          ⁡                      (                          t              -                              i                ·                τ                                      )                              ⁢              ⅆ        t            
It can be important to time-align and reduce any time mismatch between the signals e(t) and s(t−i·τ), as otherwise the tap coefficients tend to “drift” towards the first or last taps depending on the direction of the timing mismatch. This generally results in a change in the split of precursor and postcursor taps during adaptation and can result in significant “eye” opening penalties.
Conventional techniques for configuring the split of precursor and postcursor taps for an adaptive feedforward equalizer set the initial conditions on the feedforward taps appropriately. Apart from the “coefficient drift” reasons in cases of timing mismatches between the signals e(t) and s(t−i·τ) for adapting the tap coefficient ci, the regular coefficient adaptation can also result in changes in the precursor/postcursor split in the feedforward equalizer. To time-align the signals e(t) and s(t−I·τ), conventional designs set a fixed, static timing offset for the error signal. This is not sufficiently effective if the delays along the various components in the signal data path are not accurately known or if they vary with time.
Referring to FIG. 1, a conventional adaptive signal equalizer 10 includes a feedforward filter 12, an adaptive coefficients generator 14 and an output signal slicer 16. Additionally, if decision feedback equalization is desired, a feedback filter 20 further filters the final output signal 17 from the slicer 16 to provide a feedback signal 21 which is combined in a signal combiner 22 (e.g., signal summing circuit) with the initially equalized signal 13 provided by the feedforward filter 12. The resulting equalized signal 13/23 is sliced by the signal slicer 16 to produce the output signal 17.
An additional signal combining circuit 18 combines the input 13/23 and output 17 signals of the slicer 16 to provide the error signal 19 representing the difference between the pre-slice 13/23 and post-slice 17 signals. As is well known, this error signal 19 is processed by the adaptive coefficients generator 14, along with the incoming data signal 11, to produce the adaptive coefficients 15 for the feedforward filter 12.
Additionally, so as to compensate for internal signal delays ts, tc within the feedforward filter 12 and signal slicer 16, signal delay circuits 24s, 24e can be included in the signal paths for the incoming data signal 11 and pre-slice signal 13/23.
Referring to FIG. 2, a conventional feedforward filter 12 processes the incoming data signal 11 to produce the equalized signal 13 using a series of signal delay elements 32, multiplier circuits 34 and output summing circuit 36 in accordance with well-known techniques. Each of the successively delayed versions 33a, 33b, . . . , 33n, as well as the incoming data signal 11, is multiplied in one of the multiplication circuits 34a, 34b, 34c, . . . , 34n with its respective adaptive coefficient signal 15a, 15b, . . . , 15n. The resulting product signals 35a, 35b, . . . , 35n are summed in the signal summing circuit 36, with the resulting sum signal forming the equalized signal 13.
Referring to FIG. 3, a conventional adaptive coefficients generator 14 processes the incoming data signal 11 and feedback error signal 19 using a series of signal delay elements 42, signal multipliers 44 and signal integrators (e.g., low pass filters) 46 in accordance with well known techniques. The incoming signal 11 is successively delayed by the signal delay elements 42a, 42b, . . . , 42n to produce successively delayed versions 43a, 43b, . . . , 43n of the incoming signal 11. Each of these signals 11, 43a, 43b, . . . , 43n is multiplied in its respective signal multiplier 44a, 44b, . . . , 44n with the feedback error signal 19. The resulting product signals 45a, 45b, . . . , 45n are individually integrated in the signal integration circuits 46a, 46b, . . . , 46n to produce the individual adaptive coefficient signals 15a, 15b, . . . , 15n. 